The Archdiocese of Southwark built this year's Diocesan Day for Catechists around the theme "Truth and Beauty in Catechesis in the Age of AI", and gave the keynote to Matthew Harvey Sanders. He explains in plain language how these systems are built and why that makes them fluent and wrong in the same sentence, draws three cautions out of Magnifica Humanitas, and offers four ways to use the tools in a parish alongside five things he'd never hand to one. He also leaves catechists three tests to put to any tool before it reaches a parish. Read the full talk below.

I. A Technologist's Account

This Easter, both of London's archdioceses had more adults preparing to enter the Church than in any year since 2011, and in this archdiocese half of them were thirty-five or younger. You'll know far better than I do what that has asked of your parishes. At the same moment, something else has arrived. For the first time, anyone with a phone can put a question about the faith, at any hour, to a system that has read more theology than any of us will in a lifetime, and have a fluent answer in seconds. Never before have so many people been asking to come in at the very moment a machine has offered to answer them.

I build those systems. My company, Longbeard, makes Magisterium AI and a handful of other tools that Catholics use to study their faith. I was a seminarian once, so I know a little of what your ministry asks, but I won't pretend to know it as you do. What I can offer is a technologist's account of these machines: what they're genuinely good for, where they fall short, and why. And the places where they fall short, it turns out, describe your ministry with surprising precision.

I've also sat where your candidates sit. I was baptised an Anglican, and I was received into the Catholic Church as an adult, through the RCIA. I knew very few Catholics, and I had a great many questions and concerns. A tool like ours would have helped me find the right sources and work through them. But what brought me in was the kindness and patience of the priest who led our class, and nothing I've built since has changed my mind about which of those mattered more.

Pope Leo's encyclical Magnifica Humanitas strikes the same balance. It describes AI as "a valuable tool" that, "at the same time", calls "for a measured and vigilant approach". I'd like to follow that balance this morning: to say plainly what these machines are, what they do well and where the Pope asks us to be vigilant, why all of this makes your ministry grow rather than shrink, how to use these tools well, and what it now takes to form a person. We'll finish where this morning is heading, before the Blessed Sacrament.

II. Naming the Machine

First, what these things are, in words you could use with a parent at the school gate, a Year 9 class or a sceptical headteacher.

A large language model, the kind of system behind ChatGPT, Gemini and Claude, is made in two stages.

In the first, it reads. Its makers gather an enormous library of human writing: books, articles, websites, forums, much of the public internet. The text is broken into small pieces called tokens, each a word or part of a word, and every token is turned into a long list of numbers. Then the model is set a single task, repeated trillions of times: given everything so far, predict the next token. After every round of guesses, billions of internal settings, called parameters, are nudged slightly towards the right answer. Nobody sets those by hand. After months of this, what remains is a system that is astonishingly good at continuing any piece of writing you give it.

There's a simple way to show a class what that means. Ask one of these systems the same question three times, and you'll often get three differently worded answers. At every step the model has a list of possible next words, each with a probability attached, and it chooses among them with a little deliberate randomness. Teenagers are usually surprised, and usefully so, because it shows them that the answer on the screen isn't being looked up anywhere. It's being composed, a word at a time, out of likelihoods. That's also why it can be fluent and wrong in the same sentence. It's choosing what sounds likely, and what sounds likely isn't always what's true.

Here's the part I'd most like you to be able to explain. Inside the model, words become points in a vast space of numbers, and words that are used in similar ways end up close together. "Eucharist" sits near "Mass" and "communion" and "bread" and "thanksgiving". "Grace" sits near "mercy" and "gift". The model has learnt, with extraordinary precision, where every word lives among all the others. That, in the end, is what a word means to a machine: its neighbours.

Now take the word at the centre of everything you teach. To the machine, "Eucharist" is a position, very well placed among its neighbours. To us, it's Someone. The machine knows every word that has ever stood next to that one. It has never met the One the word names. That's the whole distance between information about the faith and the faith itself, and it's the distance your ministry lives in.

Then comes the second stage. If the first is a library, the second is a finishing school. People put questions to the model, compare its replies and mark the ones they'd rather receive, and round after round it's nudged towards their choices. The industry calls this reinforcement learning from human feedback, RLHF for short, and it's where a model gets its manners: helpful, courteous, fluent. But the reply people would rather receive isn't always the true one. Researchers have found that a reply which matches a person's own views is more likely to be chosen, so unless its makers take deliberate care, a model learns to flatter.

A system like this can be cleverer at some things than any of us, and still understand nothing about the child in front of you.

III. A Valuable Tool That Requires Vigilance

The Pope begins with what these systems do well, and so should we. They "often surpass human intelligence in speed and computational capacity," he writes, "offering tangible benefits across many fields." In catechesis the need they can meet is easy to name. There aren't enough priests to sit with every enquirer at the hour their questions come, and there aren't enough catechists either. A question about the faith asked at two in the morning deserves better than whatever the internet happens to say.

A general chatbot isn't the answer to that. Left to itself, it answers from what it absorbed in training, and when that memory is thin it doesn't tell you; it produces something that sounds right. The industry calls that a hallucination. The alternative is to find the relevant texts first and have the model answer only from them, an approach called retrieval-augmented generation, or RAG. That's how we built Magisterium AI, because one of our first priorities has always been to send people back to the original sources.

Behind Magisterium AI sits a library. Over the past few years we've gathered what I believe is one of the largest online libraries of Catholic theological and philosophical works in the world: more than thirty-six thousand documents, from the Catechism and the encyclicals to Augustine's City of God, The Imitation of Christ, Teresa of Avila's Interior Castle and Julian of Norwich's Revelations of Divine Love. Ask it something, and the answer is drawn from that library, with a citation on every point that you can click to read the passage itself, in context. There's an audiobook collection too, which may reach the parent who'll never sit down with a spiritual classic but might listen to one on the school run.

So the enquirer at two in the morning can be given the Church's own teaching, in her own words, with the sources attached, in their own language. I think that's a real service to the Church. But I want to be clear about what it's for. The aim isn't to need fewer catechists because the machines have become good. It's to form more catechists, so that people come to need the machines less.

Then the Pope turns to vigilance, and names three things about our personal use of AI that "deserve careful consideration": "the ease with which results are obtained, the impression of objectivity and the simulation of human communication." Each has an engineering cause, and each points to something only a catechist can give.

The first is ease. Answers have never been cheaper, and Leo names what that can cost: "The speed and ease with which answers or summaries can be obtained risk extinguishing the desire to ask questions, which is a process that bears fruit only over time." He borrows an image from Plato for how the deepest things are learnt: people "striking upon" ideas and experiences together "like flint until the spark of understanding is kindled within us." A machine can hand a young person the answer. It can't strike the flint with her. Chesterton put the teacher's side of it in 1910: "Education is only truth in a state of transmission; and how can we pass on truth if it has never come into our hand?" The machine offers a way round his question: truth that can be handed on without ever passing through a person's hands.

A candidate who has already asked the machine what confirmation is, why the bishop anoints and what the gifts of the Spirit are arrives with a tidy paragraph on each, and has, oddly, reached the real question sooner: not "What is confirmation?" but "Why would I want it?" Ease has a quieter cost as well. A mother whose seven-year-old asks where Grandma has gone may not feel qualified, so she types the question into her phone and reads out what comes back. It will be a kind and competent paragraph. The danger isn't that the machine gives a bad answer. It's that it gives a good enough one, and she never discovers that she could have answered it herself, with her own faith, however halting.

The second is the impression of objectivity. An answer on a screen feels neutral, as if it came from nowhere. It didn't. The Pope observes that these systems "reflect the cultural assumptions of those who designed and trained them", and from the inside I can tell you how that happens: through the writing the model learnt from in the first place, the people who rated its answers in the second stage, and the values its makers wrote down. In 1938 Fulton Sheen wrote that there were "not over a hundred people in the United States who hate the Catholic Church", but "millions, however, who hate what they wrongly believe to be the Catholic Church." A model trained on much of the public internet has read a great deal of that caricature alongside the real thing, and on its own it can't always tell them apart. That's the strongest argument I know for answering from the Church's own sources.

There's a further limit that matters for your work. The model doesn't know who's asking. It sees only the words in front of it, what the industry calls its context window. The same question about sin, typed by a scrupulous twelve-year-old and by a sixth-former trying to catch you out, gets much the same answer, because the model can't see a face. A catechist can. What looks like objectivity is often just an answer written for nobody in particular.

The third is the simulation of human communication, and it's the one I'd watch most closely with teenagers. A model that has learnt to flatter makes a convincing companion, and there are now apps built entirely around one: an AI that plays the part of a friend, a boyfriend or a girlfriend, and is never too busy to talk. This summer ministers said the ones sold as "romantic companion" chatbots "will have to enforce a minimum age of 18." Governments don't set age limits on things nobody is using. Here the Pope sees further than most of my industry does: "the danger is not so much that a person may believe they are communicating with another person, but rather that they may gradually lose the very desire to form genuine human connections."

Gaudium et Spes calls conscience "the most secret core and sanctuary of a man. There he is alone with God, Whose voice echoes in his depths." Hold that beside a teenager's chat window. A sanctuary is where you're alone with God. The chat window is the opposite on both counts. It's rarely private, because what's typed into it is usually stored on a company's servers, and it's never silent, because it always has something to say back. And it has been trained to say what she'd like to hear. A confidant who never disappoints her isn't forming her conscience. It's taking its place.

Behind all three cautions lies something simpler, which the Pope states without hedging. These systems "do not undergo experiences, do not possess a body, do not feel joy or pain, do not mature through relationships and do not know from within what love, work, friendship or responsibility mean." It's the word "Eucharist" again: perfectly placed among its neighbours, by something that has never met the One it names. A machine can describe the faith with great accuracy. It has no body to bring to anyone, and the faith has always been handed on by people who bring theirs: to a classroom, a hospital bed, a parish hall, a font.

IV. Why This Ministry Grows Rather Than Shrinks

You might expect someone from my industry to tell you that the machines will make catechists less necessary. I think the opposite is true, and the reasons start with the size of what's coming.

Demis Hassabis, Alphabet's chief scientist and the chair of Google DeepMind, works a few miles from here in King's Cross, and in 2024 he shared the Nobel Prize in Chemistry. Speaking to The Guardian in August last year, he put the scale like this: it's "going to be 10 times bigger than the Industrial Revolution, and maybe 10 times faster." The last industrial revolution remade this country over a hundred years or more, and the Church here spent much of the nineteenth century catching up with it, building schools and parishes for the families it had drawn into the towns. Now imagine a change ten times the size, compressed into a tenth of the time.

And it isn't only software. The same intelligence is going into robots for warehouses, kitchens and care homes, and over the coming years a large share of the work people now do for a living is likely to be done by machines. Many of the people building this expect that, handled well, it will leave people with more leisure than any society in history has known, and enough to live on. Elon Musk talks about "universal high income". Pope Leo is more cautious, and writes of "a legitimate fear of a significant and rapid contraction in available jobs". Either way, a great many people are going to have far more hours in the week that no employer needs, and the young are meeting it first: the job site Adzuna reports that graduate vacancies fell to a fresh record low in July.

Then the questions come. Musk himself said in 2024 that in such a world "the question will really be one of meaning." For most of history, the struggle to make a living kept the biggest questions at arm's length. There was always a shift to work and a bill to pay. Take that away, and the questions rise to the surface. What is my life for? Does it matter, if nobody needs my work? What happens when I die? Is anyone there? A generation with the time to ask them will ask them more openly than any before it.

As a novice with the Paulist Fathers I was stationed at Berkeley, among the students of the University of California, and one thing I learnt there is that questions like these are rarely only requests for information. Underneath, they're asking whether anyone cares about their soul and wants them to become the person God had in mind. That's why I think the catechist is going to matter more in the next twenty years than at any time in living memory.

The Directory for Catechesis asks for a catechesis of "catechumenal inspiration", and from where I sit there's a striking irony in that. The world outside is busy dismantling its apprenticeships. The junior roles through which judgement and character used to be formed are the roles the machines are taking first. The Church went the other way. At the Second Vatican Council she restored the catechumenate for adults, one of the oldest apprenticeships in the West. Just as the world is losing the apprenticeship, the Church has one to offer.

Pope Leo has the right phrase for the people this asks for. He calls the digital world "a new continent to be evangelized", and says that in a particular way we need "adults to rediscover their vocation as artisans of education, prepared to work patiently each day." An artisan is precisely what a machine isn't. Artisanship is knowledge held in the hands, learnt from someone who had it, and passed on slowly, by working alongside. Elsewhere he says the same of education itself: it "is a long journey requiring patience". Nothing I build is patient. It's only tireless, which isn't the same thing.

When God wanted to be known, he didn't send us a better text. He sent his Son. "The Word became flesh and dwelt among us." And on the last night, the example he chose to leave his disciples wasn't an explanation. He washed their feet.

That's how the faith has been handed on ever since: by people whose lives make the answer believable. Saint John Henry Newman, preaching at Oxford in 1832 while still an Anglican, asked how the truth had survived at all, and answered that it had been upheld "not as a system, not by books, not by argument, nor by temporal power, but by the personal influence" of those "who are at once the teachers and the patterns of it." Teachers and patterns. A machine can be a very good teacher of facts. It can never be a pattern in Newman's sense. For a technologist there's an irony in that, because patterns are all a language model is made of, absorbed from a vast library of human writing. Newman meant something older: a life that someone else can copy. A machine has no life to offer. It can tell a young person what Christians believe. It can't show them what believing looks like. The young person who asks what their life is for won't be persuaded by a well-sourced paragraph. They'll be persuaded, if they are, by someone whose own life shows that the answer is true.

V. Putting It to Work

In practice, used well, these tools can free up time for the parts of this work that only a person can do. I'll suggest four ways to use them, three tests to put to any tool, and five things I'd never hand to a machine.

The first way is preparation rather than delivery. Imagine you're preparing a session on the gifts of the Holy Spirit for a group of confirmation candidates. You want what the Catechism teaches, what Saint Thomas Aquinas wrote about the gifts, how the rite of confirmation itself prays for them, and a story from the life of a saint that a teenager would remember, all with references you can check. With the right tool that's ten minutes' work, and the time you'd have spent searching is yours again.

Since last month, you don't have to talk to a chatbot at all. There's a feature in Magisterium AI called Search: you enter a word or a phrase, and instead of an answer you get the sources themselves, to open and read. Preparing a session on prayer, you can go straight to what the Catechism, the Fathers and the saints wrote, with no machine paraphrasing them in between. And there's one called Explore, which is the space I described earlier made visible. It's a three-dimensional map of the library in which related texts cluster together, so you can find a topic, see what sits around it, and wander into works you didn't know existed. For anyone getting to know a subject, it's the nearest thing I know to browsing the shelves of a great theological library. For now, both work in the English version.

So I'd use the three together. Explore, to get your bearings in a subject you're about to teach. Search, to read the primary texts yourself. And the chat, when you want help drafting a session or pitching something at the right level. I'd treat what comes back as a draft to pray over and change, never a script to read out, and click every citation before anything reaches a group. One user wrote to us in Spanish that, without knowing the whole Catechism and the Bible, they "feel confident in doing things right." That's the right kind of help: not a machine that does the thinking, but one that shows its working.

The second way is to bring the machine's questions into the room. Between sessions, many candidates will take their questions to a machine. So it's worth pointing them to a tool that sends them to the sources, and then opening each session by asking, "What did you ask a machine this week?" Every question that comes back into the room is one you'd otherwise never have heard, and you get to answer it face to face. When a question comes up that nobody at the table can answer, you can look it up together with Search and read the Church's own words aloud. A teenager who has only ever seen a machine's summary discovers that there's an original, and that it's usually more beautiful than the summary.

The third way is saying one truth in many ways. The same teaching has to be put differently to a twelve-year-old, to an adult in the RCIA, to a sceptic and to an anxious parent. A good tool can draft several versions in a minute. Choosing the door into the truth is still your judgement, because you know who's standing in front of you. The same tools can help when a family has only just arrived, perhaps as refugees, and is still learning English. A letter about their child's preparation, translated into a language they can read, keeps them part of it while their English grows, and tells them they're welcome.

The fourth way is with the administration that surrounds every programme: the registers, the rotas, the emails to parents, the timetable for the sacramental programme, the note that the retreat has moved. None of that is the ministry, and all of it can be done faster. Every hour saved there is an hour that can go to a person.

Then there are three tests I'd put to any tool before it reaches a parish.

First, can you trust its sources? Most of the big systems can now search the internet and show you their links, which is a real step forward. But a link tells you where an answer came from, not whether the source deserves your trust, and the open internet holds the caricature Sheen described as well as the real thing. If you use a general tool, open what it cites and make sure you're reading the Church's own teaching and not someone's opinion of it. If you'd rather not do that vetting yourself, use a tool whose sources have been chosen with care. That's why we built Magisterium AI on a curated library: you still read the passage, but the vetting has been done before you ask, and that saves you time. And if a tool can't show you where an answer came from at all, it isn't fit for preparing catechesis, however fluent it sounds.

Second, is it built to end the conversation or to prolong it? Almost nobody asks this, and it comes straight from how technology companies measure themselves. Most consumer apps are judged by a number called retention: how many people come back tomorrow, and how long they stay. A product built to maximise that number will always try to keep a young person talking to it. A good catechetical tool should do the opposite. It should answer the question and then point beyond itself: to the Scriptures, to the sacraments, to the parish, to you.

Third, where does a child's question go once it's been typed? Is it stored, and for how long? Is it fed into training future models? Can someone at the company read it? Was the product designed for children, or is it merely not forbidden to them? If the people offering you the tool can't answer those questions plainly, you have your answer.

Pope Leo puts the principle behind all three in a single line: "Educating people about the use of AI, then, involves teaching them to decide when and for what purpose it ought not to be used." So there are five things I'd never hand to a machine at all.

The first is accompaniment. Walking with someone through doubt, grief or conversion can only be done by someone who is walking too.

The second is discerning readiness. Whether a young person is ready for confirmation, or an adult for baptism, is a judgement made in the Church by people who know them, never by a score.

The third is anything that stands in the place of prayer. A machine can help you prepare to pray with your group. It should never be the prayer.

The fourth is a child's confidences. What a child tells you in trust belongs to the Church's pastoral care and safeguarding, and never to a chatbot.

And the fifth is the witness itself. What hands on the faith is a person who believes it, standing in front of someone who doesn't yet.

None of this needs deciding alone. A catechetical team that spends one evening agreeing which tools it will use, for what, and where its lines are will have a clear answer when parents or the parish priest ask, and the conversation itself is formation.

VI. Forming a Person for a World Full of Machines

All of which leads to the deepest question of the morning: what does it mean to form a person in a world like this? The Pope frames it with a sentence every engineer should keep above the desk: "every technology shapes those who use it." The young people in your groups are being formed already, every day, by tools built to hold their attention. Against that, I'd name four things that now need forming.

The first is interiority. Nothing my industry makes is ever silent; a search box is always ready with something to say. So the sanctuary of conscience now has to be learnt almost as a skill: silence, an examination of conscience, a few minutes before the tabernacle with nothing to look at but the Lord. It can be simple. Two minutes of real silence at the end of a session, with nothing to look at and nothing to answer, will feel strange to teenagers the first time, and by the end of the year it may be the part they remember.

The second is attention. My industry's most refined product is the feed: a stream chosen, moment by moment, by a system that has learnt what you'll keep watching. It gives you more of what you already want. The liturgy does the opposite. The readings for next Sunday weren't chosen by anyone in this room, or by anything that has studied you, and the Church's year brings Advent and Lent whether we feel like them or not. The lectionary gives you the passages you'd have skipped. In that sense the liturgy is the anti-algorithm, and a young person formed by it learns what the feed can never teach: that what you need isn't always what you'd choose. Saint Irenaeus wrote that "our way of thinking is attuned to the Eucharist, and the Eucharist in turn confirms our way of thinking." A way of thinking can be attuned to a feed, or it can be attuned to the Eucharist. Catechesis is where that choice gets made.

The third is friction. A machine will never be disappointed in you. That sounds like a kindness, but it's really the absence of a relationship. Being known by someone who expects more of you, and who's still there when you fall short, is how people grow. A catechist who notices that a teenager has gone quiet, and says so, is offering something no app is designed to offer: a person who cares enough to be let down. Saint Josemaría Escrivá was unsentimental about what that costs. "Making a fraternal correction is always hard," he wrote. "Everyone knows this. Making fraternal corrections is the best way to help, after prayer and good example." Notice the order: prayer, then example, and only then the hard word. A machine can manage none of the three. It can't pray, it can't be an example, and the way it's trained pulls it away from the hard word.

The fourth is a dignity that was never earned. The young people in your groups are growing up in a culture that measures worth by output, and they're about to be told, by the economy and by the machines, that their output isn't needed. That's the moment the oldest thing the Church teaches becomes the most urgent: that God loved them before they could do anything at all, and will go on loving them when they can do nothing. Every baptism says that out loud: the child at the font has achieved nothing, and the whole Church rejoices over her anyway.

In its section on education, Magnifica Humanitas says that schools "are not called to follow the pace of the digital world, but to offer that which the digital sphere by itself cannot provide, namely a shared time for learning and developing trustworthy relationships." Put the word "parish" where it says "schools" and you have as good a description of catechesis in this decade as I know.

VII. Where the Gaze Goes

I began with a machine that can answer almost any question about the faith, at any hour. In the years ahead it will answer more and more of them, and some of its answers will be good. I hope ours are among them. But the questions people bring at two in the morning are rarely only questions. They're a kind of restlessness, and an answer, however good, doesn't settle it.

I work inside the industry that's about to meet that restlessness at a scale the world has never seen. And I can tell you what it's good at. It's extraordinarily good at keeping a restless heart busy: one more video, one more answer, one more message from something that never tires of you. What it has no way to do is give that heart rest, because rest isn't information. It's a Person. Near the end of Magnifica Humanitas, Pope Leo names what an age like ours needs, in one sentence: "The spirituality that we need is a Eucharistic spirituality." And the catechist, in every age, has been the one who shows people where that Person is to be found. That's the last and deepest of the machine's shortcomings, and it's the first thing a catechist is for.

I'll finish with two people who used the most powerful media of their time to hand on the faith, and who never mistook the medium for the source.

The first is Fulton Sheen, beatified in St Louis nine days ago. He was the great catechist of the broadcast age. The Catholic Hour on the radio reached an estimated four million listeners, and his television series, Life Is Worth Living, some thirty million viewers. On the eve of the beatification Pope Leo expressed the hope that his example would "inspire many others likewise to find new ways of spreading the Gospel message with joyful enthusiasm." Yet the portrait unveiled at the beatification shows him holding a piece of chalk, the chalk he used at his blackboard on television. Of everything they might have put in his hand, they chose a teacher's chalk. And every day of his priesthood he spent an hour before the Blessed Sacrament, "the hour that makes my day", as he called it. "Theological insights," he wrote, "are gained not only from the two covers of a treatise, but from two knees on a prie-dieu before a tabernacle." The machines I build can hold thousands of treatises. They have no knees.

The second is Carlo Acutis, born here in London, who taught himself to code. You'll know the website he made, cataloguing the Eucharistic miracles of the world, which became an exhibition that has travelled the world. What strikes me, as someone who builds for screens, is how he made it. He dragged his parents from one shrine to another so that he could photograph them himself. His website was made on foot. Almost everything my industry makes is designed to hold your gaze. Carlo's was built to send you somewhere else: to a church, to an altar, to a place you would have to travel to. That's my second test, passed by a teenager twenty years ago. Does it answer the question and then point beyond itself? Or does it want to be looked at?

Sheen had the broadcast. Carlo had the web. You have been handed something more powerful than either, at the very moment more adults are asking to come in than at any time in fifteen years. It may be the most powerful tool ever put in the Church's hands, and with it you can carry the Gospel further and faster than either of them could have imagined. Use it as they used theirs: boldly, and without ever mistaking the medium for the source.

In a little while you'll walk the few steps from here into the Cathedral and kneel before the Blessed Sacrament. Sheen did that for an hour every day for sixty years, and he described it in a line less well known than it deserves to be: "Sitting before the Presence is like a body exposing itself before the sun to absorb its rays." There, in front of you, will be the rest every restless heart in your parish is looking for, whether they know it or not. Every screen in their lives is competing for their gaze. The machine was built to hold it there. The Eucharist was given to lift it up.